The Mw = 5 . 6 Kanallaki earthquake of March 21 , 2020 2 in west Epirus , Greece : reverse fault model from 3 InSAR data and seismotectonic implications for 4 Apulia-Eurasia collision 5

We identify the source of the Mw = 5.6 earthquake that hit west-central Epirus on March 21 21, 2020 00:49:52 UTC. We use synthetic aperture radar interferograms tied to one permanent 22 Global Navigation Satellite System (GNSS) station (GARD). We model the source by inverting the 23 INSAR displacement data. The inversion model suggests a shallow source on a low-angle fault 24 (39°) dipping towards east with a centroid depth of 8.5 km. The seismic moment deduced from our 25 model agrees with those of the published seismic moment tensors. This geometry is compatible 26 with the Margariti thrust fault within the collision zone between Apulia and Eurasia. We also 27 processed new GNSS data and estimate a total convergence rate between Apulia and Eurasia of 8.9 28 mm yr-1 , of which shortening of the crust between the Epirus coastal GNSS stations and station 29 PAXO in the Ionian Sea is equivalent to ~ 50% of it or 4.6 mm yr−1. A 60-km wide deformation 30 zone takes up nearly most of the convergence between Apulia-Eurasia, trending N318°E. Its central 31 axis runs along the southwest coast of Corfu, along the northeast coast of Paxos, heading toward 32 the northern extremity of the Lefkada island. 33


36
The tectonics of Epirus is characterised by on-going compression due to the active collision 37 between the Apulian continental block [1,2,3,4] and the Eurasian (Aegean) plate. The Apulian 38 continental lithosphere subducts beneath Epirus and was imaged at 70-80 km depth by [5]. Both 39 seismological and geodetic data show that active deformation is occurring by crustal shortening 40 directed in a NNE-SSW to NE-SW orientation [6,7,8,9,10,11] (Fig.1). The overall tectonic strain rate 41 was estimated at 40-50 ns• yr−1 of contraction by [11] for west-central Epirus.      University of Athens (SL-NKUA). We merged the two databases by complementing arrival-time 99 data for events that are common in the two catalogues and also including unique events only

117
To improve the hypocentral distribution we applied the double-difference relocation method 118 using the algorithm HypoDD [23]. This method works by assuming that when the distance between 119 two neighbouring events is much smaller than their hypocentral distance from a station, then their 120 travel-time differences can be attributed to their inter-event distance. Relative relocation between 121 hypocentres is achieved by minimizing the double-difference between observed and calculated 122 travel-times for pairs of neighbouring events at common stations, reducing errors caused by 123 unmodeled velocity structure. To further enhance the relocation procedure, we also incorporated

140
The star represents the epicentre of the Mw=5.6 mainshock of March 21, 2020.

141
The focal mechanism solutions produced by various agencies for the Kanallaki mainshock are 142 reported in Tab. 3. The local agencies (NOA, AUTH) report centroid depths significantly lower than 143 the other agencies. Our revised moment tensor solution provided a centroid depth of 8 km. We 144 favour the east-dipping nodal plane because it fits the regional geological data (west-verging 145 thrusts; Fig. 1c; Fig. 4

148
almost pure reverse-slip. The Kanallaki region is an area of relatively low seismicity in Greece where strong earthquakes 153 are rare and occur with magnitudes M < 6.0 ( Fig. 4; see Table S1 for the record 1915-2019); however, 154 one disastrous earthquake has occurred in the 19 th century.
[24] and [25] report information on the

170
Moreover, on June 16, 1990 02:16 UTC an offshore M = 5.5 event occurred, that was studied by

171
[8] who determined a reverse focal mechanism. That earthquake was localized at 20.54°E, 39.16°N

172
(see Fig. 4), thus ~15 km to the southwest of the 2020 event with a centroid depth 8.5 ± 4 km, with 173 strike, dip and rake angles 352° ± 20°, 37° ± 9°, 110° ± 15° respectively, and a seismic moment of  We use synthetic aperture radar interferometry (InSAR) to capture the deformation produced 187 by the earthquake (Fig. 5; Fig. 6). In the broader Aegean area, InSAR is systematically used to map the signal to noise ratio of the interferograms by applying the adaptive power spectrum filter of [34] 209 with a coherence threshold of 0.3.

211
Two moderate-size aftershocks occurred during the 4-day period following the mainshock, one

235
we selected 86 values among 247, and on the ascending interferogram (see Table 5) 86 among 289. As

240
This assumption is justified by the non-detectable offset at the position time-series of the GNSS 241 station GARD (Gardiki) which is located in the SW side of the tringle (see Fig. 1 for its location). The shows that changing the zero of the interferogram by 1/8 of a fringe (3.5 mm) results in the change of 247 the moment tensor by 32%, here completely absorbed by the increase of the fault length as this is the 248 only free geometric parameter, the other free parameters being the three coordinates of the fault-top 249 centre. As our first models predicted a geodetic moment slightly lower that the seismic moment (see 250   Table S3), we use for the zero of the fringes the value that maximizes the interferogram, thus the case 251 of maximum line of sight change of 31.5 mm.

253
The ascending interferogram (Fig. 6) is noisier and therefore its zero more difficult to establish.

254
We therefore used a different method and performed the fringe tuning by aligning the geodetic 255 moment predicted for the descending interferogram with the one predicted for the ascending one.

256
The best fit is obtained when using 34 mm for the displacement on the top quarter of fringe of the    scatter between data and model is 3.3 mm for the 86-descending data (Fig. 8 top) and 3.9 mm for the 282 86-ascending data (Fig. 8 bottom). We assume that the seismic fault plane is the one with low-angle 283 reverse fault dipping towards ENE. This is the most consistent with the geomorphology and the 284 tectonic context of the area.

286
We invert for the 3D coordinates of the fault-top centre and for the fault length. We do not 287 invert for the strike, dip and rake angles, assuming 329°, 39° and 102° respectively, i.e. using the 288 average values of the various determinations of the focal mechanism parameters (see Tab. 3). The

289
dip angle is, among the three angles, the one that is less constrained. For the two other angles there is 290 a trade-off between strike and rake. We tested models with the strike and/or the rake free but we 291 could not constrain well those parameters. This was expected given the almost circular shape of the 292 interferograms, especially the ascending one (Fig. 6).

294
In a first series of inversions we find that the width of the fault has the tendency to become 295 large, between 4.5 and 6 km and the length small, between 4 and 5 km, with the surface area 296 remaining almost constant. As the width is not well constrained, we fix it at 4.7 km, a value the leads 297 to an equal length of 4.7 km, thus a square fault. We also released the amplitude of slip and rake: we 298 found that the amplitude had the tendency to grow (i.e. from an initial 0.3 m to 0.6 m or more) and 299 the rake to remain steady. In the second set of inversions we constrained the slip to be 0.5 m, as the 300 standard scaling laws (i.e. [36]) do not suggest a higher value for this earthquake, given also the size     . 10a) and Loutsa that imply the existence of a major thrust there. As the Margariti thrust fault 350 (see Fig. 4) is more than 12-km long its down-dip extension could reach the Kanallaki area (Fig. 10),

351
~10 km to the east, and such a geometry complies also with the geodetic centroid depth of 8.5 km 352 that we obtained for the 2020 seismic fault.

354
To the northwest of Kanallaki and west of Gardiki (Fig. 10a), the Lippa mountain forms a ~400  Table S4 for coordinates). b): simplified section (not to scale) of the crustal structure beneath the

Tectonic strain and the determination of the deformation zone west of Kanallaki
We analyzed new GNSS data of the Greek permanent stations in Epirus and north Ionian Sea (Corfu and Paxoi islands; Tab. 8). The GNSS stations belong to the NOANET geodetic network antenna calibrations, random walk troposphere estimation, and the FES2004 ocean loading model.

378
The tectonic velocities and uncertainties retrieved from the analysis of the time series are in Tab. 8.

392
Using the velocities of Tab. 8 we calculate the convergence towards Apulia in the azimuth 393 N228°Ε which is the azimuth that fit best the vectors of the stations located close to the coast (thus 394 excluding the two stations of Ioannina). We note that the orientation of the coast is globally 407 408

411
By modelling this convergence with a simple back-slip model in elastic half-space (Fig. 11) we 412 find the best fit for a total convergence rate of 8.9 mm yr -1 , thus an extra compression of 3.7 mm yr -1 413 that is recovered in the regions located to the east and north of Ioannina. The modelled central axis 414 of the convergence band, with azimuth N318°E, passes along the southwest coast of Corfu, along the 415 northeast coast of Paxos, heading toward the northern extremity of the Lefkada island (Fig. 12). The 416 extent of the deforming area (box in Fig. 12), when defined as the band that accommodates 80% of 417 the total convergence, is 60 km, with a best fitting locking depth of 11 km. Unfortunately, the limited 418 surface displacement data are not dense enough to allow for a formal inversion so as to recover the 419 slip deficit patterns along the deformation front. Therefore, PAXO appears to be located already slightly on the Apulian side of the deforming axis of the plate boundary zone with Africa passing close to both islands and continuing along the coast of Albania. Between Corfu and Epirus (the Igoumenitsa stations IGOU -HGOU) the data and 429 the model (Fig. 11) show that there is no significant shortening within the velocity uncertainties of ± 430 0.2 mm yr -1 , confirming the conclusions of [42}.

432
It is noteworthy that there is a 120 ns yr -1 active shortening of the crust between the two GNSS Because of the concern for Ioannina we made Sentinel differential 460 interferograms immediately after that event but we saw no deformation. This is because the 461 moderate-size event was too deep. Also, GPS data processing showed no displacement at the two 462 stations of Ioannina (Fig. 1).   for the Kanallaki earthquake that were not used in the inversion due to low quality and noise, Figure S5: Aerial 502 overview of the earthquake epicentre area,